I am conducting an analysis among single (no spouse or partner) female/male headed families with children under 18. I was trying to understand how IPUMS creates the HHTYPE variable and have been comparing it to the PUMS HHT variable. When using the IPUMS HHTYPE variable, I limit to only head of households (relate == 1) and the age of youngest child to be under 18 years old (ygnch (0-17) to obtain my sample of interest.
When using PUMS HHT I similarly limit HHT to heads of household (RELSHIPP== 2) own children in household under 18 (HUPAOC 1-3), people who are not married with spouse present (MSP does not equal to 1 or missing), and people who are not in unmarried partnerships (CPLT does not equal to 3/4). I find that my observations (using 2023 ACS data) are off by 20-150 for single women and men with own children under 18.
Are there other variables being used to create HHTYPE or should HHT not be used as the base variable when trying to match PUMS and IPUMS?
It sounds like you are interested in understanding discrepancies between estimates of households that are headed by single-persons who have a child under the age of 18 using the IPUMS HHTYPE variable and the Census Bureau HHT variable (among others). I can think of a few circumstances under which HHTYPE and HHT would not match, but the examples for your specific use-case are pretty limited. It would be very helpful to have identifiers for specific households (e.g., YEAR and SERIAL or CBSERIAL combinations) where you are seeing differences between the HHTYPE and HHT variables for your population of interest. I have included two (rather involved!) examples of households that might be measured differently by these two variables as well as some generic info on HHTYPE below.
- Households containing a householder who is married, no one with a relationship value of spouse (RELATE =2 in IPUMS), but a married “Other relative” (RELATE = 10) who is the only other person in the household who reports being married. IPUMS would link the householder and the other relative as being spouses via the SPLOC variable, and count any children of either partner towards the total family size. I am unsure how Census Bureau would handle this (i.e., would they prioritize marital status or the lack of a “spouse” in the household).
- Households with an unmarried householder and multiple unmarried partners. IPUMS would not assign a SPLOC value in this case as our algorithm can only assign one SPLOC; I am unsure what the Census Bureau CPLT variable would assign in this case. If the householder has a child, this may be another source of discrepancy.
Broadly, HHTYPE classifies households based on marital status (MARST), SEX, and family size; currently, it is designed to mirror the “family type” concept reported in the 2000 decennial census documentation (see "Family Type on page B-16 of the 2000 PUMS codebook). It uses the variable FAMSIZE for family size, which is based on the IPUMS family interrelationship algorithm that identifies a person’s probable coresident spouse (SPLOC) and parent(s) (MOMLOC1, MOMLOC2, POPLOC1, POPLOC2). If two people are linked via SPLOC and one of them is linked as a parent to the child, the spouse is also linked as a parent to that child. Using example one from my list above, I have constructed a toy household to show how this might work:
|RELATE | MARST | AGE | SEX|
|Householder | Married, spouse present | 47 | M|
|Other relative | Married, spouse present | 45 | F|
|Other relative | Never married | 10 | M|
The adult other relative would be linked as the mother of the child other relative; the adult other relative would also be linked as the spouse of the householder because there are no other married persons in the household, and the householder would then be linked as the child other relative’s father.
Note that the IPUMS family interrelationship variables do not use self-reported values of whether or not a person has children in the household (as these data or the line number/household location of one’s own children is not available in many datasets). To the extent that IPUMS links children to parents where the Census Bureau does not or links people as spouses when the Census Bureau does not, the family size and composition will be different between the IPUMS and the Census Bureau family definitions. While the IPUMS family definitions identify cohabiting partners as members of the same family and identify unrelated (aka secondary) families within a household that are unrelated to the householder, this is unrelated to your specific question as you are interested in single people and because HHTYPE is assigned to 9 if the householder is linked to an unmarried partner via SPLOC (in most cases – there are a handful of exceptions).
I look forward to some specific examples so I can (hopefully) provide a more concrete answer!